Developing a claims’-based algorithm to identify US patients with COPD

Background: Claims databases are useful sources to design algorithms for care management strategies. However, for conditions like COPD that depend on clinical data (e.g., spirometry) for accurate diagnosis, their non-availability in claims implies that any claims’ based identification algorithm that is developed will need to be validated in order to minimize potential misclassification. Aim and Objective: To develop and validate a predictive model to identify COPD patients using claims data integrated with medical records. Methods: A predictive model was developed using US administrative claims from the HealthCore Integrated Research Database linked to spirometry results data of 2,005 patients with a claims-based diagnosis of COPD between 01/01/12–11/30/13. LASSO regression with 10-fold cross-validation was used to model 1,505 COPD cases (post-bronchodilator FEV1/FVC Results: The final model included 14 predictors, the strongest of which included: presence of ≥3 COPD diagnostic codes each 30 days apart, ≥1 pulmonary rehabilitation claim, COPD hospitalization count, and ≥1 claim for smoking. The model had good discrimination in identifying COPD cases (c-statistic 0.73) and a high cross-validated positive predictive value (0.86, 95%CI=0.84-0.88) for a probability threshold of ≥75% to identify COPD cases. Conclusions: Using multiple diagnostic codes in combination with COPD related healthcare utilization data were found to be valid parameters for identifying COPD patients using claims data alone.